Scalar indexing on GPU arrays in elements initialization

[huiyuxie]

This issue was found in #117, where the scalar indexing on GPUs happens in the function init_elements when we try to cache data on GPUs.

This issue can be resolved by create a struct for LobattoLegendreBasisGPU to gain finer control of data types within the struct. (i.e., In this way, we can keep the nodes on CPUs, instead of forcing it to be on GPUs like other arrays.)